10x your Productivity with Custom Cursor MCP Agents

mCP Hands-On Tutorial Summary

Key Concepts:

• mCP (Model Context Protocol): Universal USB port for AI, enabling AI agents to interact with external systems.
• Cloe Desktop: Cloe running as a local application.
• Cursor: Code editor with AI integration.
• Windsurf: Another platform for AI integration.
• fastmCP: A library for creating mCP servers.
• Yahoo Finance: A library for retrieving stock prices.
• Neon: Serverless PostgreSQL database.
• smithery.gl.a: A website to find existing mCPs.

1. Introduction to mCP

• mCP is described as a "Universal USB port for AI," allowing AI agents to connect to external systems and significantly enhance their capabilities.
• A basic AI agent can read/write code, create/delete files, and run terminal commands. With mCP, it can access databases, APIs, private data, and interact with applications.
• The video highlights the evolution from simple chatbots to single agents, then multiple agents, and now multiple agents with external tools via mCP.
• mCP addresses the limitations of isolated code generation by enabling access to browser consoles, network tabs, asset generation, and API integration.

2. Creating a Custom mCP Server (Stock Price Example)

• The video demonstrates creating a custom mCP server to retrieve stock prices using Yahoo Finance.
• Step-by-step process:

  1. Install required packages: pip install mCP-cli Yahoo-finance

  2. Create a file named app.py

  3. Import necessary libraries: import yfinance as yf and from mCP_server.fastmCP import fastMC CP

  4. Initialize fastMC CP: mCP = fastMC CP(name="Stock Prices")

  5. Define a function to get the stock price:

     def get_stock_price(ticker_symbol: str) -> float:
         ticker = yf.Ticker(ticker_symbol)
         current_price = ticker.fast_info.last_price
         return current_price
     

  6. Register the function as an mCP tool:

     @mCP.tool
     def get_stock_price(ticker_symbol: str) -> float:
         ticker = yf.Ticker(ticker_symbol)
         current_price = ticker.fast_info.last_price
         return current_price
     

  7. Run the mCP server:

     if __name__ == "__main__":
         mCP.run(transport="stdio")
     

  8. In Cloe Desktop, edit the configuration file (edit config in the developer settings).

  9. Add a new entry with the path to the app.py file and the Python command (obtained using which python).

  10. Restart Cloe Desktop.

• The example uses the yfinance library to fetch real-time stock prices.
• The presenter emphasizes the simplicity of creating an mCP server using this method.

3. Integrating with Cloe Desktop

• After creating the mCP server, the video shows how to integrate it with Cloe Desktop.
• The configuration involves specifying the path to the Python executable and the app.py script in the Cloe Desktop settings.
• Once configured, Cloe Desktop can use the mCP tool to answer questions like "Get the stock price of Apple."
• The video demonstrates how to grant permission for the chat to access the mCP tool.

4. Integrating with Cursor

• The video explains how to integrate existing mCP servers (like the Neon database mCP) into Cursor.
• It references websites like smithery.gl.a for finding existing mCPs.
• Step-by-step process:
  1. Obtain the mCP command from a source like smithery.gl.a (requires an API key for Neon).
  2. Open Cursor settings and navigate to the mCP section.
  3. Add a new mCP server, providing a name (e.g., "Neon") and the command.
  4. Save the configuration.
• Cursor then displays the available functions from the mCP server (e.g., "list projects").
• The video demonstrates using the "list projects" function to retrieve project details from a Neon database.

5. Integrating with Windsurf

• The video briefly touches on integrating mCP with Windsurf.
• Windsurf has an mCP icon for configuration.
• The configuration process is similar to Cloe Desktop and Cursor, involving specifying the command for the mCP server.
• After configuration and a refresh, Windsurf can access the mCP tools.
• The example shows Windsurf using the Neon mCP to "list the branches in Neon."

6. Neon Database Integration

• The video showcases a use case of integrating with a Neon serverless PostgreSQL database.
• Neon automatically shuts down when not in use, saving costs.
• With mCP, users can create applications that interact with their database directly from Cursor, including creating schemas, getting database status, and saving data.

7. Conclusion

• The video provides a basic example of creating a custom mCP and integrating it with different IDEs.
• The presenter encourages viewers to explore mCP further and suggests watching another video on creating custom mCPs for a more complete understanding.
• The presenter will put all the code in the description below for the viewers to try out.

Notable Quotes:

• "mCP is nothing but an Universal USB port for AI."
• "Previously it's limited code generation isolated from your development environments unable to access external resources but with m CP you are able to access browser console logs Network tabs generate assets integrate with apis connect to any service you need."
• "With just one bit of command you are able to create a custom mCP."
• "I haven't seen any other easy method to create mCP and this is the easiest way to implement mCP."

Main Takeaways:

• mCP is a powerful tool for extending the capabilities of AI agents by connecting them to external systems.
• Creating custom mCP servers is relatively simple using the fastmCP library.
• mCP can be integrated with various IDEs and platforms, including Cloe Desktop, Cursor, and Windsurf.
• Real-world applications include accessing databases, APIs, and other services directly from within a development environment.